Cloudflare’s AI Search service has reached general availability, introducing robust improvements in multimodal search capabilities, including embedding image pixels directly and integrated optical character recognition for PDFs. The service expands file size limits and implements a new pricing model while maintaining a free tier on all Workers plans.
- Native image embeddings integrated with text for faster, richer visual search
- OCR support adds scanned PDF text extraction up to 10 MiB file size
- New billing starts November 2026; free tier available on all Workers plans
Infrastructure signal
AI Search leverages Cloudflare’s Workers AI, Vectorize, R2, and Browser Run services to provide a fully managed pipeline for indexing and retrieval. The introduction of native image embeddings using Matryoshka Representation Learning balances representation detail with storage and query efficiency, a critical infrastructure advance enabling fast multimodal searches without ballooning costs or latency.
Supporting scanned PDFs through integrated optical character recognition (OCR) also enhances data ingestion pipelines by enabling text extraction from image-only documents, expanding the range of searchable content. The support for files up to 10 MiB accommodates larger and more complex datasets, but also signals a shift in resource utilization that will influence cloud storage and processing demands.
Developer impact
Developers gain access to improved multimodal search capabilities that work with any chat model, including direct image queries and combined image-text queries. The ability to embed images natively in a shared vector space with text improves the precision and relevance of search results, reducing workarounds previously needed with caption-only approaches.
The enhanced file support and OCR also mean developers can now include scanned documents and a wider range of file formats in their searchable indexes without preprocessing outside the platform. This simplifies workflows, accelerates deployment, and reduces the need for separate OCR services. However, developers will need to incorporate the new billing considerations starting November 2026, balancing volume with the free tier limits.
What teams should watch
Infrastructure and platform teams should monitor the impact of multimodal embeddings and larger ingestion sizes on storage capacity and query latency within their deployments. Careful tuning may be required to optimize costs given the new pricing structure tied to AI Search ingestion tokens, especially when using OCR and handling image-heavy queries.
Observability teams will want to enhance monitoring of query patterns and resource consumption to anticipate scaling needs and cost control. Teams responsible for API integrations and developer experience should communicate new feature capabilities and billing changes clearly to maintain adoption momentum and manage developer expectations.
Finally, product and search-specialist teams should explore the potential unlocked by native image retrieval and OCR to improve user-facing search relevancy and usability, especially for applications relying on product discovery, document search, and visual data navigation.